Comput Manag Sci (2012) 9:205–231

DOI 10.1007/s10287-011-0133-z

ORIGINAL PAPER

Supply chain network operations management

of a blood banking system with cost and risk
minimization

Anna Nagurney · Amir H. Masoumi · Min Yu

Received: 12 November 2010 / Accepted: 3 August 2011 / Published online: 13 August 2011
© Springer-Verlag 2011

Abstract Blood service operations are a key component of the healthcare system
all over the world and yet the modeling and the analysis of such systems from a com-
plete supply chain network optimization perspective have been lacking due to their
associated unique challenges. In this paper, we develop a generalized network opti-
mization model for the complex supply chain of human blood, which is a life-saving,
perishable product. In particular, we consider a regionalized blood banking system
consisting of collection sites, testing and processing facilities, storage facilities, dis-
tribution centers, as well as points of demand, which, typically, include hospitals. Our
multicriteria system-optimization approach on generalized networks with arc multi-
pliers captures many of the critical issues associated with blood supply chains such as
the determination of the optimal allocations, and the induced supply-side risk, as well
as the induced cost of discarding the waste, while satisfying the uncertain demands as
closely as possible. The framework that we present is also applicable, with appropriate
modifications, to the optimization of other supply chains of perishable products.

Keywords Supply chains · Perishable products · Blood banking · Healthcare ·

Medical waste · Network optimization · Cost minimization · Risk minimization ·
Multicriteria decision-making · Generalized networks · Variational inequalities

1 Introduction

In today’s world, supply chains are more complex than ever before. Consumers’
demand for new products as well as the still-critical economic situation require that

A. Nagurney (B) · A. H. Masoumi · M. Yu

Department of Finance and Operations Management, Isenberg School of Management,
University of Massachusetts, Amherst, MA 01003, USA
e-mail: nagurney@gbfin.umass.edu

123
206 A. Nagurney et al.

companies, as well as organizations, be more innovative while also becoming more

cost-effective in the procurement and production of their products and services as well
as in their delivery. However, despite numerous significant achievements, the disci-
pline of supply chain management (SCM) is still incapable of satisfactorily addressing
many practical, real-world challenges (Georgiadis et al. 2005).
Supply chains for time-sensitive products and, in particular, for perishable products,
pose specific and unique challenges. By definition, a perishable product has a limited
lifetime during which it can be used, after which it should be discarded (Federgruen
et al. 1986). Examples of perishable goods include food and food products, medicines
and vaccines, cut flowers, etc. (see, e.g., Hsu et al. 2007; Zanoni and Zavanella 2007;
Osvald and Stirn 2008; Ahumada and Villalobos 2009). Clearly, not all perishable
products are alike and, notably, in some cases, such as that of medicines and vaccines,
the quality of a product, or lack thereof, may result in a matter of “life or death” for
its consumers.
In this paper, we focus on a specific perishable product—that of human blood—and
the optimization of a blood banking network system. Nahmias (1982) claimed that:
“The interest among researchers in perishable inventory problems has been sparked
primarily by problems of blood bank management. Some of the possible reasons for
this interest might be that blood bank research has been supported by public funds.”
Prastacos (1984), subsequently, provided a review, to that date, of blood inventory
management, from both theoretical and practical perspectives. Whether or not Nah-
mias’ statement is still valid—considering all the recent concerns about the safety of
perishable products, blood bank management from a supply chain network perspec-
tive merits a fresh and updated approach. This topic is especially timely today, since
it has been reported that the number of disasters and the number of people affected by
disasters has been growing over the past decade and blood is certainly a life-saving
product (cf. Nagurney and Qiang 2009).
Blood service operations are a key component of the healthcare system all over
the world. According to the American Red Cross, over 39,000 donations are needed
everyday in the United States, alone, and the blood supply is frequently reported to
be just 2 days away from running out. Of 1,700 hospitals participating in a survey
in 2007, a total of 492 reported cancellations of elective surgeries on one or more
days due to blood shortages. While for many hospitals, the reported number of blood-
related delays was not significant, hospitals with as many days of surgical delays
as 50 or even 120 have been observed. Furthermore, in 2006, the national estimate
for the number of units of whole blood (WB) and all components outdated by blood
centers and hospitals was 1,276,000 out of 15,688,000 units (Whitaker et al. 2007).
Considering also the ever-increasing hospital cost of a unit of red blood cells with a
6.4% increase from 2005 to 2007 further highlights the criticality of this perishable,
life-saving product. In the US, this criticality has become more of an issue in the
Northeastern and Southwestern states since this cost is meaningfully higher compared
to that of the Southeastern and Central states. Moreover, hospitals were responsible
for approximately 90% of the outdates, with this volume of medical waste imposing
discarding costs to the already financially-stressed hospitals (The New York Times
2010).

123
Supply chain network operations management of a blood banking system 207

Several authors have applied integer optimization models such as facility loca-
tion, set covering, allocation, and routing to address the optimization/design of supply
chains of blood or other perishable critical products (see Jacobs et al. 1996; Pierskalla
2004; Yang 2006; Sahin et al. 2007; Sivakumar et al. 2008; Cetin and Sarul 2009;
Ghandforoush and Sen 2010). In addition, inventory management methods (cf. Cohen
and Pierskalla 1979; Karaesmen et al. 2011, and the references therein), Markov mod-
els (Boppana and Chalasani 2007) as well as simulation techniques (Rytila and Spens
2006; Katsaliaki and Brailsford 2007; Mustafee et al. 2009) have been used to handle
blood banking systems. Yegul (2007), in his dissertation, which has extensive refer-
ences on the subject of blood supply chains, also utilized simulation for a blood supply
chain with a focus on Turkey. He noted that there were few studies which consider
multiple echelons (as the model in our paper does). Haijema et al. (2007) used a Mar-
kov dynamic programming and simulation approach with data from a Dutch blood
bank (see also the dissertation of Haijema 2008).
In this paper, we develop a multicriteria system-optimization framework for a
regionalized blood supply chain network, such as that of the American Red Cross.
In the US, the American Red Cross (ARC) is the major supplier of blood products
to hospitals and medical centers satisfying over 45% of the demand for blood com-
ponents nationally (Walker 2010). A system-optimization approach is believed to be
mandated for critical supplies (Nagurney et al. 2011) in that the demand for such
products must be satisfied as closely as possible at minimal total cost. The use of
a profit maximization criterion, as in Nagurney (2010a,c), is not appropriate for an
organization such as the American Red Cross, due to its non-profit status.
Unlike many other supply chain models that assume a fixed lifetime for a perish-
able good (see, e.g., Hwang and Hahn 2000; Omosigho 2002; Zhou and Yang 2003),
our system-optimization approach for supply chain network management, as we shall
demonstrate, captures the perishability/waste that occurs over the relevant links associ-
ated with various activities of the supply chain, similar to the spatial price equilibrium
model in Nagurney and Aronson (1989). However, in contrast to the latter model, here
we also take into account the discarding cost of the waste over the relevant links as
well as the discarding cost of outdated product at the demand points due to the possible
excess supply delivered. Furthermore, we capture in the model the uncertainty of the
demand and the associated shortage penalties at the demand points. System-optimi-
zation models have been developed to capture various issues of SCM including that
of mergers and acquisition as well as network design (cf. Nagurney 2009, 2010b,c;
Nagurney and Woolley 2010; Nagurney et al. 2011; Nagurney and Nagurney 2010).
This paper is organized as follows. In Sect. 2, we describe, in detail, the structure
of a regionalized blood banking system. We develop the supply chain network model
for the blood banking system problem, and establish that the multicriteria optimiza-
tion problem is equivalent to a variational inequality problem, with nice features for
computations. We also present simple numerical examples and conduct sensitivity
analysis. Our model has several novel features:

1. it captures perishability of this life-saving product through the use of arc multi-
pliers;
2. it contains discarding costs associated with waste/disposal;

123
208 A. Nagurney et al.

3. it handles uncertainty associated with demand points;

4. it assesses costs associated with shortages/surpluses at the demand points, and
5. it quantifies the supply-side risk associated with procurement.
In Sect. 3, we propose an algorithm which, when applied, yields the optimal level of
blood product flows. We then apply the algorithm to compute the solution to a larger-
scale numerical example using data motivated by a real-world application in order to
further illustrate the modeling and computational framework for blood banking sup-
ply chain network management. In Sect. 4, we summarize the results and present our
conclusions.

2 The supply chain network model of a regionalized blood banking system

In this section, we develop the supply chain network model for regionalized blood
banks. It is necessary to mention that our model for blood banking management is
applicable to many perishable products, with minor modifications, but with the same
foundations. Also, it is worth noting that although the structure of the network for
a blood banking system, or the way that the modules of the supply chain are called,
may be slightly different from country to country, or from one region to another, our
network framework is sufficiently general to address any blood supply chain network.

2.1 The components of a regionalized blood banking system

In most parts of the world, blood banking operations systems conduct procurement and
distribution in a regionalized manner. In other words, there exists a Regional Blood
Center in each geographic area which is in charge of coordination and administra-
tion of its lower-level units. Nevertheless, despite advances in storage and distribution
technologies, hospitals may need to acquire blood products from suppliers that are
located in other regions, sometimes even hundreds of miles away.
In the US, for example, the regional divisions of the American Red Cross over-
see the entire operation of their corresponding regions. Other suppliers of blood are
hospitals—typically the larger ones with blood collection programs—which, however,
account for less than 5% of the market share (Whitaker et al. 2007). There also exist
private blood suppliers across the country.
Since 1960, the Red Cross has been reimbursed by the hospitals for the costs asso-
ciated with providing blood to hospital patients. The Red Cross does not charge for the
blood itself; it only recovers the costs associated with the recruitment and screening
of potential donors, the collection of blood by trained staff, the processing and testing
of each unit of blood in state-of-the-art laboratories, and the labeling, storage, and
distribution of blood components.
Figure 1 depicts a network topology of a regionalized blood banking system as for
the ARC in the US. In this network, the top level (origin) node represents the ARC
regional division. Every other node in the network denotes a component/facility in
the system. A path connecting the origin node to a destination node, corresponding to
a demand point, consists of a sequence of directed links which correspond to supply

123
Supply chain network operations management of a blood banking system 209

Fig. 1 Supply chain network topology for a regionalized blood bank

chain network activities that ensure that the blood is collected, processed, and, ulti-
mately, distributed to the demand point. We assume that in the supply chain network
topology there exists at least one path joining node 1 with each destination node. This
assumption guarantees that the demand at each demand point will be met as closely as
possible, given that we will be considering uncertain demand for blood at each demand
point. The solution of the model yields the optimal flows of blood at minimum total
cost and risk, as we shall demonstrate.
In the network in Fig. 1, we assume that the division is considering n C blood collec-
tion sites constituting the second tier of the network. Many of these collection sites are
mobile or temporary locations while others are permanent sites. In the case of drastic
shortages; i.e., natural or man-made disasters, the regional divisions are likely to need
to import blood products from other regions or even other countries, an aspect that is
excluded from this model. In the current topology, the first set of the links connecting
the origin node to the second tier corresponds to the process of “blood collection.”
These collection sites are denoted by: C S1 , C S2 , . . . , C Sn C S .
The next set of nodes, located in the third tier, are the blood centers. There exist
n BC of these facilities in one region, denoted by BC1 , BC2 , . . . , BCn BC , to which the
whole blood (WB) is shipped after being collected at the collection sites. Thus, the
next set of links connecting tiers two and three of the network topology represents
the “shipment of collected blood.”
The fourth tier of the network is composed of processing facilities, commonly
referred to as component labs. The number of these facilities in one region is assumed

123
210 A. Nagurney et al.

to be n C L . These facilities are denoted by C L 1 , . . . , C L n C L , respectively, and are

typically located within the blood center locations. At these labs, the collected blood
is separated into parts, i.e., red blood cells (RBC) and plasma, since most recipients
need only a specific component for transfusions. Every unit of donated whole blood—
450 to 500 ml on average—can provide one unit of RBCs and one unit of plasma.
What we refer to as the flow of product is the amount of WB on the first three sets of
links. Likewise, the flow on the links thereafter denotes the number of units of RBCs
processed at the component labs which are, ultimately, delivered to the hospitals.
We exclude plasma and other side derivatives from our model for several reasons.
First of all, although plasma can be derived from donated WB, in practice, plasma is
mainly produced in a different process called apheresis. Apheresis is a blood dona-
tion method where the blood is passed through an apparatus that separates out one
particular constituent—plasma and returns the remainder—red blood cells—to the
donor. Secondly, plasma can be stored frozen for an extended period of time, typically
one year, which is not comparable to the approximately 5-week lifetime of RBCs.
Most important of all, WB and RBCs account for the major part of donations and
transfusions rather than plasma and other components (Whitaker et al. 2007).
The safety of the blood supply is considered to be the most important issue in blood
services. In the US, federal law mandates that every single unit of donated blood
be tested before being transfused, regardless of the number of the times one donor
has donated blood in the past. The National Testing Laboratories of the American
Red Cross are in charge of this vital task, testing blood for multiple infectious dis-
ease markers, including but not limited to HIV, hepatitis, and the West Nile Virus
(Redcrossblood.org 2010). These facilities are owned and operated by the ARC, and
require heavy investments for specialized equipment. Presently, only 5 testing labs
are operating across the US, and these labs are shared among 36 blood regions. Only
a small sample of every donated blood unit is sent to the testing labs, overnight, and
these samples are discarded regardless of the results of the tests. Due to the high per-
ishability of many of the blood products, the two processes of testing and separating
take place concurrently yet sometimes hundreds of miles away. If the result of a test for
a specific unit of donated blood at the testing lab turns out to be positive, the remainder
of that unit will be later discarded at the corresponding storage facility. In our model,
the set of the links connecting the component labs to storage facilities corresponds to
“testing and processing,” and the cost on these links represents the operational cost of
testing and processing combined. The fraction of the flow lost during or as a result of
the testing process is also taken care of in our model.
The fifth set of nodes denotes the short-term storage facilities. There are n S F of such
nodes in the network, denoted by S F1 , S F2 , . . . , S Fn S F , which are usually located in
the same place as the component labs. The links connecting the upper level nodes to
the storage facilities denote the procedure of “storage” of the tested and processed
blood before it is shipped to be distributed.
The next set of nodes in the network represents the distribution centers, denoted by
DC1 , DC2 , . . . , DCn DC , where n DC is the total number of such facilities in the region.
Distribution centers act as transshipment nodes, and are in charge of facilitating the
distribution of blood to the ultimate destinations. The links connecting the storage tier
to the distribution centers are of “shipment” link type.

123
Supply chain network operations management of a blood banking system 211

The last set of links joining the bottom two tiers of the network are “distribution”
links, ending in n R demand points. Hospitals and surgical medical centers are the
predominant users of blood. The actual but uncertain demands of the demand points
R1 , R2 , . . . , Rn R are denoted by: d R1 , d R2 , . . . , d Rn R , respectively.
It is necessary to mention that specific components of the system may physically
coincide with some others; however, this network topology is process-based rather than
location-based, which is compatible with our blood banking problem. Moreover, as
mentioned earlier, in general cases of perishable product supply chains, these facilities
may be located far apart which can be nicely addressed using our presented model.
The supply chain network topology is represented by G = [N , L], where N and L
denote the sets of nodes and links, respectively. The ultimate solution of the complete
model will yield the optimal flow on the various links of the network.

2.2 The formulation

Our formulation is of a single-period type, where the time horizon spans the various
activities of procurements, processing, and distribution. Since WB is highly perish-
able, all modules of the blood supply chain network tend to avoid long term storage
(except for plasma, which is excluded from our model). Hence, the assumption of
a single-period time horizon is realistic with the focus of this paper being on oper-
ations management, rather than on inventory management. Nevertheless, our model
takes into account the potential shortage associated with the uncertain demand at the
demand points, that is, the lost “sales.” In addition, the surplus penalty can address
additional relevant costs, whether in terms of excess supply or even if short-term
inventory holding cost is included.
Associated with each link of the network is a unit operational cost function that
reflects the cost of operating the particular supply chain activity, that is, the collection
of blood at blood drive sites, the shipment of collected blood, the testing and pro-
cessing, the storage, and the distribution. We denote these links by a, b, etc. The unit
operational cost on link a is denoted by ca and is a function of flow on that link, f a .
The total operational cost on link a is denoted by ĉa , and is constructed as:

ĉa ( f a ) = f a × ca ( f a ), ∀a ∈ L . (1)

The link total cost functions are assumed to be convex and continuously differentiable.
The origin/destination (O/D) nodes consist of the pairs of nodes (1, Rk ); 1, . . . , n R ,
where Pk denotes the set of paths, which represent alternative associated possible
supply chain network processes, joining (1, Rk ). P denotes the set of all paths joining
node 1 to the destination nodes, and n P denotes the number of paths.
Let vk denote the projected demand for blood at the demand point Rk ; k =
1, . . . , n R . We assume that the demand at each demand point is uncertain with a
known probability distribution. Recall that dk denotes the actual demand at demand
point Rk ; k = 1, . . . , n R , and is a random variable with probability density func-
tion given by Fk (t). Let Pk be the probability distribution function of dk , that is,
D
Pk (Dk ) = Pr ob(dk ≤ Dk ) = 0 k Fk (t)d(t). Therefore,

123
212 A. Nagurney et al.

−
k ≡ max {0, dk − vk }, k = 1, . . . , n R , (2)
+
k ≡ max {0, vk − dk }, k = 1, . . . , n R , (3)

where − +
k and k represent the shortage and surplus of blood at demand point Rk ,
respectively.
The expected values of the shortage (− +
k ) and the surplus (k ) are given by:

∞
E(−
k ) = (t − vk )Fk (t)d(t), k = 1, . . . , n R , (4)
vk
vk
E(+
k )= (vk − t)Fk (t)d(t), k = 1, . . . , n R . (5)
0

Due to the vitalness of the availability of blood at the demand points, a relatively
large penalty of λ− k is associated with the shortage of a unit of blood at demand point
Rk , where λ− k corresponds to the social cost of a death or a severe injury of a patient,
due to a blood shortage. Also, since blood is extremely perishable and will be outdated
if not used over a certain period after being delivered, the outdating penalty of λ+ k is
assigned to the unit of a possible supply surplus. Note that, in our formulation, this
surplus penalty is charged to the organization even though the ARC is not directly
responsible for the outdated blood at the hospitals once it is delivered to them. This
is because human blood is scarce, and the ARC aims to minimize the amount of out-
dated blood at demand points, which actually dominates the amount of blood waste
during the other activities of blood banking within the ARC network (Rios 2010).
Hence, λ+ k , in the case of blood (as for other perishable products), includes the cost of
short-term inventory holding (cold storage), and, possibly, the discarding cost of the
outdated product. It is necessary to mention that having excessive supplies outdated
at the demand points not only imposes a discarding cost on the already financially
stressed healthcare institutions such as hospitals, but also leads to further environmen-
tal damage. Similar examples of penalty costs, due to excessive supplies, as well as to
shortages, can be found in the literature (see, e.g., Dong et al. 2004; Nagurney et al.
2011). These penalties can be assessed by the authority who is contracting with the
organization to deliver the blood.
Thus, the expected total penalty at demand point k; k = 1, . . . , n R , is:

E(λ− − + + − − + +
k k + λk k ) = λk E(k ) + λk E(k ). (6)

Nevertheless, the demand points (such as hospitals) are not the only modules of the
blood supply chain in which the perishability of the collected blood occurs. Through-
out the processes of blood collection, shipment, testing and processing, storage, and
distribution, a fraction of the collected blood may deteriorate, be lost, or be wasted. The
fraction of the lost product depends on the type of the activity since various processes
of collection, testing, storage, etc., lead to different amounts of waste. This fraction,
in general, can also be different among the various facilities at the same tier of the

123
Supply chain network operations management of a blood banking system 213

network, depending upon the technology used, the efficiency of the staff personnel,
and so forth.
Hence, we associate with every link a in the network, a multiplier αa , which, for
all activities of the blood supply chain, lies in the range of (0,1] where αa = 1 means
that αa × 100% of the initial flow on link a reaches the successor node of that link,
reflecting that there is no waste/loss on link a. The average percentage of loss due to
the testing process was reported to be 1.7% (Sullivan et al. 2007); consequently, the
corresponding multiplier, αa , would be equal to 1 − 0.017 = 0.983.
As mentioned earlier, f a denotes the (initial) flow on link a. Let f a denote the final
flow on that link; i.e., the flow that reaches the successor node of the link. Therefore,

f a = αa f a , ∀a ∈ L . (7)

Thus, the waste/loss on link a, denoted by wa , is equal to:

wa = f a − f a = (1 − αa ) f a , ∀a ∈ L . (8)

The organization such as ARC is responsible for discarding this waste which is
potentially hazardous. Contractors are typically employed to remove and dispose of
the waste. The corresponding discarding cost, ya , is a function of the waste, wa , which
is charged to the organization:

 
ya (wa ) = ya ( f a − f a ) = ya (1 − αa ) f a , ∀a ∈ L . (9a)

Since αa is constant, and known a priori, a new total discarding cost function, ẑ a ,
can be defined accordingly, which is a function of the flow, f a , and is assumed to be
convex and continuously differentiable:

ẑ a = ẑ a ( f a ), ∀a ∈ L . (9b)

Also, let x p represent the (initial) flow of blood (or a general perishable product)
on path p joining the origin node with a destination node. The path flows must be
nonnegative, that is,

x p ≥ 0, ∀ p ∈ P, (10)

since the product will be collected, shipped, etc., in nonnegative quantities.

Let μ p denote the multiplier corresponding to the throughput on path p, which is
defined as the product of all link multipliers on links comprising that path, that is,


μp ≡ αa , ∀ p ∈ P. (11)
a∈ p

123
214 A. Nagurney et al.

The projected demand at demand point Rk , vk , is the sum of all the final flows on
paths joining (1, Rk ):

vk ≡ x pμp, k = 1, . . . , n R . (12)
p∈Pk

Indeed, although the amount of blood that originates on a path p is x p , the amount
(due to perishability) that actually arrives at the destination of this path is x p μ p .
We also define the multiplier, αap , which is the product of the multipliers of the
links on path p that precede link a in that path. This multiplier can be expressed as:
⎧ 
⎪
⎪ δap
⎨ αa  , if {a  < a} = Ø,
αap ≡ a  <a (13)
⎪
⎪
⎩
δap , if {a  < a} = Ø,

where {a  < a} denotes the set of the links preceding link a in path p, δap is defined
as equal to 1 if link a is contained in path p, and 0, otherwise, and Ø denotes the null
set. Hence, αap is equal to the product of all link multipliers preceding link a in path
p. If link a is not contained in path p, then αap is set to zero. If a belongs to the first
set of links, the blood collection links, this multiplier is equal to 1. The relationship
between the link flow, f a , and the path flows is as follows:

fa = x p αap , ∀a ∈ L . (14)
p∈P

Similar examples of multipliers corresponding to the loss/waste on links or paths

can be found in the literature (see, e.g., Nagurney and Aronson 1989).
We group the path flows into the vector x. Also, the link flows, and the projected
demands are grouped into the respective vectors f and v.
The total cost minimization objective faced by the organization includes the total
cost of operating the various links, the total discarding cost of waste/loss over the links,
and the expected total blood supply shortage cost as well as the total discarding cost
of outdated blood at the demand points. This optimization problem can be expressed
as:

  
nR
 − + 
Minimize ĉa ( f a ) + ẑ a ( f a ) + λk E(− +
k ) + λk E(k ) (15)
a∈L a∈L k=1

subject to: constraints (10), (12), and (14).

As mentioned earlier, the minimization of total costs is not the only objective of
suppliers of perishable goods. One of the most significant challenges for the ARC,
for example, is to capture the risk associated with different activities in the blood
supply chain network. Unlike the demand which can be projected according to the
historical data, albeit with some uncertainty involved, the amount of donated blood at

123
Supply chain network operations management of a blood banking system 215

the collection sites has been observed to be highly stochastic. Even though the ARC
encourages blood donors to make appointments beforehand, donors may miss their
appointments due to inclement weather situations, traffic, personal issues, etc.
Interestingly, disasters, such as the 2010 earthquake in Haiti, may stimulate people’s
sympathy and dramatically increase the number of blood donors. As in Nagurney et al.
(2005), we introduce a total risk function r̂a corresponding to link a for every blood
collection link. This function is assumed to be convex and continuously differentiable,
and a function of the flow, that is, the amount of collected blood, on its corresponding
link. The organization attempts to minimize the total risk over all links connecting the
first two tiers of the network, denoted by L 1 ⊂ L. The remainder of the links in the
network, i.e., the shipment of collected blood, the processing, the storage, shipment,
and the distribution links, comprise the set L C C
1 . The subset L 1 and its complement L 1
partition the entire set of links L, that is, L 1 ∪ L 1 = L.
C

Thus, the risk minimization objective function for the organization can be
expressed as:

Minimize r̂a ( f a ), (16)
a∈L 1

where r̂a = r̂a ( f a ) is the total risk function on link a.

The supply chain network optimization problem for a regionalized blood banking
system can be expressed as a multicriteria decision-making problem. The organiza-
tion seeks to determine the optimal levels of blood processed on each supply chain
network link subject to the minimization of the total cost (operational and discarding)
as well as the minimization of the total supply risk. We associate with the total supply
risk objective, (16), a risk aversion factor θ , which is assigned by the decision-maker.
Thus, the multicriteria optimization problem is:

  
nR
 − 
+ 
Minimize ĉa ( f a ) + ẑ a ( f a ) + λk E(−
k ) + λ +
k E( k ) + θ r̂a ( f a )
a∈L a∈L k=1 a∈L 1
(17)

subject to: constraints (10), (12), and (14).

The above optimization problem is in terms of link flows. It can also be expressed
in terms of path flows:

  nR
 − 
+ 
Minimize Ĉ p (x) + Ẑ p (x) + λk E(− +
k ) + λk E(k ) + θ R̂ p (x)
p∈P k=1 p∈P
(18)

subject to: constraints (10) and (12), where the total operational cost, Ĉ p (x), the total
discarding cost, Ẑ p (x), and the total risk, R̂ p (x), corresponding to path p are, respec-
tively, derived from C p (x), Z p (x), and R p (x) as follows:

123
216 A. Nagurney et al.

Ĉ p (x) = x p × C p (x), ∀ p ∈ P, (19a)

Ẑ p (x) = x p × Z p (x), ∀ p ∈ P, (19b)
R̂ p (x) = x p × R p (x), ∀ p ∈ P, (19c)

with the unit cost functions on path p, i.e., C p (x), Z p (x), and R p (x), in turn,as below:

C p (x) ≡ ca ( f a )αap , ∀ p ∈ P, (20a)
a∈L

Z p (x) ≡ z a ( f a )αap , ∀ p ∈ P, (20b)
a∈L

R p (x) ≡ ra ( f a )αap , ∀ p ∈ P. (20c)
a∈L 1

Next, we present some preliminaries that enable us to express the partial derivatives
of the expected total shortage and discarding costs of outdated blood at the demand
points solely in terms of path flow variables. Observe that, for each O/D pair wk :

∂ E(− k ) ∂ E(− k ) ∂vk

= × , ∀ p ∈ Pk ; k = 1, . . . , n R . (21)
∂xp ∂vk ∂xp

By Leibniz’s integral rule, we have:

⎛∞ ⎞
 ∞
∂ E(− ) ∂ ∂
k
= ⎝ ⎠
(t − vk )Fk (t)d(t) = (t − vk )Fk (t)d(t)
∂vk ∂vk ∂vk
vk vk
= Pk (vk ) − 1, k = 1, . . . , n R . (22a)

Therefore,
⎛ ⎞
∂ E(−
k )

= Pk ⎝ x p μ p ⎠ − 1, k = 1, . . . , n R . (22b)
∂vk
p∈Pk

On the other hand, we have:

∂vk ∂ 
= x pμp = μp, ∀ p ∈ Pk ; k = 1, . . . , n R . (23)
∂xp ∂xp
p∈Pk

The above is correct since the μ p ’s are constant values. Therefore, from (22b) and
(23), we conclude that
⎡ ⎛ ⎞ ⎤
∂ E(− ) 
k
= μ p ⎣ Pk ⎝ x p μ p ⎠ − 1⎦ , ∀ p ∈ Pk ; k = 1, . . . , n R . (24)
∂xp
p∈Pk

123
Supply chain network operations management of a blood banking system 217

Similarly, for the surplus, we have:

∂ E(+ k ) ∂ E(+ k ) ∂vk

= × , ∀ p ∈ Pk ; k = 1, . . . , n R , (25)
∂xp ∂vk ∂xp
⎛v ⎞
k vk
∂ E(+ ) ∂ ∂
k
= ⎝ (vk − t)Fk (t)d(t)⎠ = (vk − t)Fk (t)d(t)
∂vk ∂vk ∂vk
0 0
(26a)
= Pk (vk ), k = 1, . . . , n R .

Thus,
⎛ ⎞
∂ E(+ ) 
k
= Pk ⎝ x pμp⎠ , k = 1, . . . , n R . (26b)
∂vk
p∈Pk

From (26b) and (23) we have:

⎛ ⎞
∂ E(+ ) 
k
= μ p Pk ⎝ x pμp⎠ , ∀ p ∈ Pk ; k = 1, . . . , n R . (27)
∂xp
p∈Pk

Let K denote the feasible set such that:

nP
K ≡ {x|x ∈ R+ }. (28)

Before deriving the variational inequality formulation of the problem, we establish

a lemma that formalizes the construction of the partial derivatives of the path total
operational cost, the total discarding cost, and the total risk with respect to a path flow.

Lemma 1 The partial derivatives of the total operational cost, the total discarding
cost, and the total risk with respect to the corresponding path flow are, respectively,
given by:
 
∂ q∈P Ĉq (x)  ∂ ĉa ( f a )
= αap , ∀ p ∈ P, (29a)
∂xp ∂ fa
a∈L
 
∂ q∈P Ẑ q (x)  ∂ ẑ a ( f a )
= αap , ∀ p ∈ P, (29b)
∂xp ∂ fa
a∈L
 
∂ R̂
q∈P q (x)  ∂ r̂a ( f a )
= αap , ∀ p ∈ P. (29c)
∂xp ∂ fa
a∈L 1

123
218 A. Nagurney et al.

Proof We establish the equivalence for (29a); the equivalences (29b) and (29c) can be
obtained in a similar fashion. The partial derivative of the total operational cost with
respect to the flow on path p is first defined as:
 
∂ q∈P Ĉq  ∂ Ĉq
= , ∀ p ∈ P, (30a)
∂xp ∂xp
q∈P

which, based on the total path cost (19a), can be rewritten as:
 
∂ q∈P Ĉq  ∂(Cq xq )  ∂Cq
= = Cp + xq , ∀ p ∈ P. (30b)
∂xp ∂xp ∂xp
q∈P q∈P

According to the definition of Cq (x) in (20a), we have:


∂ ca αaq
∂Cq a∈L
 ∂ca  ∂ca ∂ f a
= = αaq = αaq , ∀ p, q ∈ P. (31)
∂xp ∂xp ∂xp ∂ fa ∂ x p
a∈L a∈L

On the other hand, by referring to (14), yields:

∂ fa
= αap , ∀a ∈ L , ∀ p ∈ P. (32)
∂xp

From (31) and (32), we obtain:

∂Cq  ∂ca
= αap αaq , ∀ p, q ∈ P. (33)
∂xp ∂ fa
a∈L

Substituting (33) into (30b) yields:

 
∂ q∈P Ĉq   ∂ca  ∂ca
= Cp + xq αap αaq = C p + xq αap αaq .
∂xp ∂ fa ∂ fa
q∈P a∈L a∈L q∈P

Thus,
 
∂ q∈P Ĉq  ∂ca 
= Cp + αap xq αaq , ∀ p ∈ P. (34)
∂xp ∂ fa
a∈L q∈P

123
Supply chain network operations management of a blood banking system 219

By substituting proper equivalences from (14) and (20a) into (34), we have:
 
∂ q∈P Ĉ q   ∂ca  ∂ca

= ca αap + αap f a = ca + f a αap , ∀ p ∈ P.
∂xp ∂ fa ∂ fa
a∈L a∈L a∈L
(35)

On the other hand, from (1):

∂ ĉa ∂ca
= ca + fa , ∀a ∈ L . (36)
∂ fa ∂ fa

From (35) and (36), we conclude that:

 
∂ q∈P Ĉ q  ∂ ĉa ( f a )
= αap , ∀ p ∈ P. (37)
∂xp ∂ fa
a∈L

Thus, (29a) has been established.


We now derive the variational inequality formulation of the blood supply chain
network optimization problem in terms of path flows and link flows.
Theorem 1 The vector x ∗ is an optimal solution to the multicriteria optimization
problem (18), subject to (10) and (12), if and only if it is a solution to the variational
inequality problem: determine the vector of optimal path flows x ∗ ∈ K , such that:
⎡     ⎛ ⎞
∂ ∗ ∗
 q∈P Ĉ q (x ) ∂ q∈P Ẑ q (x )
nR  
⎣ + + λ+ ⎝ x ∗p μ p ⎠
k μ p Pk
∂xp ∂xp
k=1 p∈Pk p∈Pk
⎛ ⎛ ⎞⎞  ⎤
∂ ∗
 q∈P R̂q (x )
− ⎝
−λk μ p 1 − Pk ⎝ ∗
x pμp ⎠ ⎠ +θ ⎦
∂xp
p∈Pk
×[x p − x ∗p ] ≥ 0, ∀x ∈ K . (38)

The variational inequality (38), in turn, can be rewritten in terms of link flows as:
determine the vector of optimal link flows, and the vector of optimal projected demands
( f ∗ , v ∗ ) ∈ K 1 , such that:
  ∂ ĉa ( f ∗ ) ∂ ẑ a ( f ∗ ) ∂ r̂a ( f a∗ )

a
+ a
+θ × [ f a − f a∗ ]
∂ fa ∂ fa ∂ fa
a∈L 1
  ∂ ĉa ( f ∗ ) ∂ ẑ a ( f ∗ ) 
+ a
+ a
× [ f a − f a∗ ]
C
∂ f a ∂ f a
a∈L 1


nR
 + 
+ λk Pk (vk∗ ) − λ− ∗ ∗
k (1 − Pk (vk )) × [vk − vk ] ≥ 0, ∀( f, v) ∈ K 1 , (39)
k=1

123
220 A. Nagurney et al.

where K 1 denotes the feasible set as defined below:

K 1 ≡ {( f, v)|∃x ≥ 0, (10), (12), and (14) hold}.

Proof First, we prove the result for path flows (cf. (38)).
The convexity of Ĉ p , Ẑ p , and R̂ p for all paths p holds since ĉa , ẑ a , and r̂a were
assumed to be convex for all links a. We need to verify that λ− − +
k E(k ) + λk E(k )
+

is also convex. We have:

∂2  − + 
λk E(− +
k ) + λk E(k )
∂xp 2

∂ 2 E(− k ) + ∂ E(k )
2 +
= λ− + λ , ∀ p ∈ Pk ; k = 1, . . . , n R . (40a)
k
∂ x p2 k
∂ x p2

Substituting the first order derivatives from (24) and (27) into (40a) yields:
⎡ ⎛ ⎞ ⎤
∂2  −  ∂ 
λ E(− + +
k ) + λk E(k ) = λk
−
μ p ⎣ Pk ⎝ x p μ p ⎠ − 1⎦
∂ x p2 k ∂xp
p∈Pk
⎛ ⎞
∂ 
+ λ+ μ p Pk ⎝ x pμp⎠
k
∂xp
p∈Pk
⎛ ⎞

= (λ− +
k + λk )(μ p ) Fk
2 ⎝ x p μ p ⎠ > 0, ∀ p ∈ Pk ; k = 1, . . . , n R . (40b)
p∈Pk

The above inequality holds provided that (λ− +

k + λk ), i.e., the sum of shortage and
surplus penalties, is assumed to be positive. Hence, λk E(−
− + +
k ) + λk E(k ), and, as a
consequence, the multicriteria objective function in (18) is also convex.
Since the objective function (18) is convex and the feasible set K is closed and
convex, the variational inequality (38) follows from the standard theory of variational
inequalities (see Nagurney 1999).
As for the proof of the variational inequality (39), now that (38) is established, we
can apply Lemma 1. Also, from (12) and (14), we can rewrite the formulation in terms
of link flows and projected demands rather than path flows. Thus, the second part of
Theorem 1, that is, the variational inequality in link flows (39), holds.


Note that variational inequality (38) can be put into standard form (see Nagurney
1999) as follows: determine X ∗ ∈ K such that:

F(X ∗ )T , X − X ∗  ≥ 0, ∀X ∈ K, (41)

where ·, · denotes the inner product in n-dimensional Euclidean space, X ∈ R n , and
F(X ) is an n-dimensional function from K to R n , with F(X ) being continuous and
the feasible set K being closed and convex.

123
Supply chain network operations management of a blood banking system 221

Indeed, if we define the feasible set K ≡ K , and the column vector X ≡ x, and
F(X ), such that:
⎡  
∂ Ĉq (x) ∂ Ẑ q (x) ⎛ ⎞
⎢ q∈P 
⎢ q∈P
F(X ) ≡ ⎢ + + λ+
k μ p Pk
⎝ x pμp⎠
⎣ ∂xp ∂xp
p∈Pk

 ⎤
⎛ ⎛ ⎞⎞ ∂ R̂q (x)
 ⎥
q∈P ⎥
−λ− ⎝
k μ p 1 − Pk
⎝ x p μ p ⎠⎠ + θ ; p ∈ Pk ; k = 1, . . . , n R⎥ ,
∂xp ⎦
p∈Pk

(42)

then the variational inequality (38) can be reexpressed in standard form (41).
We will utilize variational inequality (38) in path flows for our computations since
our proposed computational procedure will yield closed form expressions at each iter-
ation. Once we have solved problem (38), by using (14), which relates the links flows
to the path flows, we can obtain the solution f ∗ that minimizes the total cost as well
as the total supply risk (cf. (17)) associated with the optimization of the supply chain
network of a regionalized blood banking system.

2.3 Illustrative blood supply chain network numerical examples

In order to further illustrate the above model, we now present several simple examples.
Consider the blood supply chain network topology in Fig. 2 in which the organiza-
tion has a single blood collection site, a single blood center, one component lab, one
storage facility, a single distribution center and is to serve a single demand point. The
links are labeled as in Fig. 2, that is, a, b, c, d, e, and f .

Example 1 The total cost functions on the links were:

ĉa ( f a ) = f a2 + 6 f a , ĉb ( f b ) = 2 f b2 + 7 f b , ĉc ( f c ) = f c2 + 11 f c ,

ĉd ( f d ) = 3 f d2 + 11 f d ,
ĉe ( f e ) = f e2 + 2 f e , ĉ f ( f f ) = f f2 + f f .

We assumed that there was no waste so that αa = 1 for all links in Fig. 2. Hence,
all the functions ẑ a were set equal to 0 for all the links a, . . . , f .
The total risk cost function on the blood collection link a was: r̂a = 2 f a2 , and the
risk aversion factor, θ , was assumed to be 1.
There is only a single path p1 which was defined as: p1 = (a, b, c, d, e, f ) with
μ p1 = 1.
We assumed that the demand for the product followed a uniform distribution on the
x
interval [0, 5] so that: P1 (x p1 ) = 5p1 .

123
222 A. Nagurney et al.

Fig. 2 Supply chain network topology for numerical examples 1 and 2

The penalties were: λ− +

1 = 100, λ1 = 0.
+ −
Substitution of the values of λ1 , λ1 , μ p1 , and θ into (38), yields:

∂ Ĉ p1 (x ∗ ) ∂ R̂ p1 (x ∗ )
− 100(1 − P1 (x ∗p1 )) + × [x p1 − x ∗p1 ] ≥ 0, ∀x ∈ K .
∂ x p1 ∂ x p1
(43)

Under the assumption that x ∗p1 > 0, the left-hand side of inequality (43) must be
equal to zero, that is:

∂ Ĉ p1 (x ∗ ) ∂ R̂ p1 (x ∗ )
− 100(1 − P1 (x ∗p1 )) + = 0. (44)
∂ x p1 ∂ x p1

It follows from Lemma 1 that:

∂ Ĉ p1 (x ∗ ) ∂ ĉa ( f a∗ ) ∂ ĉb ( f b∗ ) ∂ ĉc ( f c∗ ) ∂ ĉd ( f d∗ )

= αap1 + αbp1 + αcp1 + αd p1
∂ x p1 ∂ fa ∂ fb ∂ fc ∂ fd
∂ ĉe ( f e∗ ) ∂ ĉ f ( f f∗ )
+ αep1 + α f p1 . (45)
∂ fe ∂ff

Since αap1 = αbp1 = αcp1 = αd p1 = αep1 = α f p1 = 1, and f a∗ = f b∗ = f c∗ =

f d∗ = f e∗ = f f∗ = x ∗p1 , with substitution into (45), gives us:

123
Supply chain network operations management of a blood banking system 223

∂ Ĉ p1 (x ∗ )
= (2 f a∗ + 6) + (4 f b∗ + 7) + (2 f c∗ + 11) + (6 f d∗ + 11)
∂ x p1
+(2 f e∗ + 2) + (2 f f∗ + 1) = 18x ∗p1 + 38. (46)

Similarly:

∂ R̂ p1 (x ∗ ) ∂ r̂a ( f a∗ )
= αap1 = 4 f a∗ = 4x ∗p1 . (47)
∂ x p1 ∂ fa

Therefore, using the above relationships, (44) may be reexpressed as:

 
x ∗p1
18x ∗p1 + 38 − 100 1 − + 4x ∗p1 = 0, (48)
5

whose solution yields the optimal path flow: x ∗p1 = 1.48, and the corresponding opti-
mal link flow pattern: f a∗ = f b∗ = f c∗ = f d∗ = f e∗ = f f∗ = 1.48. Following (12), the
projected demand is equal to: v1∗ = x ∗p1 = 1.48.
Example 2 Example 2 had the same data as Example 1 except that now there was a
loss associated with the testing and processing link with αc = .8. Hence, we now set
(cf. (9a, b)) ẑ c = .5 f c2 and μ p1 = αc = .8.
Similar to the solution procedure used for Example 1, from variational inequality
formulation (38), under the assumption that x ∗p1 > 0, the following equation must
hold for Example 2:

∂ Ĉ p1 (x ∗ ) ∂ Ẑ p1 (x ∗ ) ∂ R̂ p1 (x ∗ )
+ − 100 × 0.8(1 − P1 (0.8 × x ∗p1 )) + = 0. (49)
∂ x p1 ∂ x p1 ∂ x p1

Since in this example, αap1 = αbp1 = αcp1 = 1, αd p1 = αep1 = α f p1 = 0.8, f a∗ =

f b∗ = f c∗ = x ∗p1 , and f d∗ = f e∗ = f f∗ = 0.8x ∗p1 , therefore:

∂ Ĉ p1 (x ∗ )
= (2 f a∗ + 6) + (4 f b∗ + 7) + (2 f c∗ + 11)
∂ x p1
+0.8(6 f d∗ + 11) + 0.8(2 f e∗ + 2) + 0.8(2 f f∗ + 1) (50)
= 8x ∗p1 + 24 + 0.8(10 × 0.8x ∗p1 + 14) = 14.4x ∗p1 + 35.2.

Also,

∂ Ẑ p1 (x ∗ ) ∂ ẑ c ( f c∗ )
= αcp1 = f c∗ = x ∗p1 . (51)
∂ x p1 ∂ fc

The partial derivative of the total risk function was equal to that of Example 1:

∂ R̂ p1 (x ∗ ) ∂ r̂a ( f a∗ )
= αap1 = 4 f a∗ = 4x ∗p1 . (52)
∂ x p1 ∂ fa

123
224 A. Nagurney et al.

However, now we have

0.8x ∗p1
P1 (0.8x ∗p1 ) = . (53)
5
Therefore, the following equation needs to be solved:
 
0.8x ∗p1
14.4x ∗p1 + 35.2 + x ∗p1 − 80 1 − + 4x ∗p1 = 0. (54)
5

The new optimal path flow solution was: x ∗p1 = 1.39, which corresponds to the
optimal link flow pattern: f a∗ = f b∗ = f c∗ = 1.39, and f d∗ = f e∗ = f f∗ = 1.11. The
projected demand was: v1∗ = x ∗p1 μ p1 = 1.11. Comparing the results of Examples 1
and 2 reveals the fact that when perishability is taken into consideration, with αc = .8
and the above data, the organization chooses to produce/ship slightly smaller quanti-
ties so as to minimize the discarding cost of the waste, despite the shortage penalty of
λ− 1.
Note that when λ− ∗
1 = 200, the optimal path flow solution becomes: x p1 = 2.77,
and the corresponding optimal link flow pattern: f a = f b = f c = 2.77, and f d∗ =
∗ ∗ ∗

f e∗ = f f∗ = 2.22, with a projected demand of: v1∗ = x ∗p1 μ p1 = 2.22.

In fact, using Eq. (49), with λ−1 substituted for 100, we can derive the optimal path
flow x ∗p1 as a function of λ−1 , that is:

100(λ−
1 − 44)
x ∗p1 = . (55)
16λ−
1 + 2425

Therefore, an appropriate increase in the unit shortage penalty cost λ− 1 results in

the organization processing larger volumes of the blood product, even exceeding the
optimal flow in Example 1, which makes sense intuitively. Furthermore. for λ− 1 ≤ 44,
the organization acquires and, hence, processes and distributes, zero units of the blood
product.

Sensitivity analysis

We conducted additional sensitivity analysis by varying the loss associated with the
testing and processing activity, αc , and the unit shortage penalty cost, λ−1 , respectively.
The computed optimal path flows x ∗p1 and the optimal values of the objective function
(OF) (18) are reported in Table 1.
It is interesting to note from Table 1 that, under a specific unit penalty cost, the path
flow will be reduced when the loss of testing and processing, (1−αc ), increases within
some specific range. For instance, when λ− 1 is 200 and the loss, 1 − αc , increases from
0.4 to 0.8, the optimal path flow decreases from 2.83 to 0.88. However, the optimal
value of the objective function always keeps on increasing. Table 1 also illustrates,
under the same loss associated with testing and processing, the optimal path flow rises
with an increase in the unit shortage penalty cost, λ− 1.

123
Supply chain network operations management of a blood banking system 225

Table 1 Computed optimal path flows x ∗p1 and optimal values of the objective function as αc and λ−
1 vary

λ−
1 αc

.2 .4 .6 .8 1

100 x ∗p1 0.00 0.58 1.16 1.39 1.48

OF 250.00 246.96 234.00 218.83 204.24
200 x ∗p1 0.88 2.40 2.83 2.77 2.61
OF 494.19 439.52 376.23 326.94 288.35
300 x ∗p1 2.10 3.74 3.86 3.54 3.20
OF 715.12 581.15 464.85 387.17 331.44
400 x ∗p1 3.20 4.76 4.57 4.03 3.55
OF 914.75 689.71 525.36 425.56 357.63
500 x ∗p1 4.20 5.57 5.09 4.37 3.79
OF 1096.03 775.55 569.30 452.16 375.23
1000 x ∗p1 8.09 7.95 6.41 5.19 4.33
OF 1799.11 1027.94 681.89 515.88 415.67
2000 x ∗p1 12.69 9.80 7.27 5.68 4.65
OF 2631.32 1224.45 755.64 554.47 439.05
3000 x ∗p1 15.33 10.58 7.60 5.86 4.76
OF 3107.51 1307.25 783.73 568.57 447.39
4000 x ∗p1 17.03 11.01 7.77 5.96 4.82
OF 3415.88 1352.89 798.54 575.88 451.68

Furthermore, the optimal path flow in Example 2, assuming a nonnegative value,

as a function of the link multiplier and the shortage penalty simultaneously can be
expressed as follows:

5αc (λ−
1 − 14) − 120
x ∗p1 = . (56)
αc2 (λ−
1 + 50) + 65

3 The algorithm and an additional numerical example

In this section, we recall the Euler method, which is induced by the general iterative
scheme of Dupuis and Nagurney (1993). Its realization for the solution of the blood
bank SCM problem governed by variational inequality (38) (see also (41)) induces
subproblems that can be solved explicitly and in closed form.
Specifically, at an iteration τ of the Euler method (see also Nagurney and Zhang
1996) one computes:

X τ +1 = PK (X τ − aτ F(X τ )), (57)

where PK is the projection on the feasible set K and F is the function that enters the
variational inequality problem (41).

123
226 A. Nagurney et al.

As shown in Dupuis and Nagurney (1993); see also Nagurney and Zhang (1996), for
convergence of the general iterative scheme, which
induces the Euler method, among
other methods, the sequence {aτ } must satisfy: ∞ τ =0 aτ = ∞, aτ > 0, aτ → 0,
as τ → ∞. Specific conditions for convergence of this scheme can be found for a
variety of network-based problems, similar to those constructed here, in Nagurney
and Zhang (1996) and the references therein. Applications of this Euler method to
the solution of oligopolistic supply chain network design problems can be found in
Nagurney (2010a).

Explicit formulae for the euler method applied to the blood supply chain network
variational inequality (38)

The elegance of this procedure for the computation of solutions to the blood supply
chain network operations management problem modeled in Sect. 2 can be seen in
the following explicit formulae. In particular, (57) for the blood supply chain net-
work management problem governed by variational inequality problem (38) yields
the following closed form expressions for the blood product path flows:
⎧ ⎛ ⎛ ⎛ ⎞⎞ ⎛ ⎞
⎨  
x τp+1 = max 0, x τp + aτ ⎝λ− ⎝
k μ p 1− Pk
⎝ x τp μ p⎠⎠ −λ+
k μ p Pk
⎝ x τp μ p⎠
⎩
p∈Pk p∈Pk
   !"
∂ q∈P Ĉq (x τ ) ∂ q∈P Ẑ q (x τ ) ∂ q∈P R̂q (x τ )
− − −θ ,
∂xp ∂xp ∂xp
∀ p ∈ Pwk ; k = 1, . . . , n R . (58)

This closed form expression was applied to calculate the updated product flow
during the steps of the Euler Method for our blood banking optimization problem.

Example 3 We now apply the Euler method to compute solutions to a larger-scale

numerical blood supply chain network problem. The numerical example consisted
of two blood collection sites, two blood centers, two component labs, two storage
facilities, two distribution centers, and three demand points, as depicted in Fig. 3.
We assumed that R1 was a small surgical center while R2 and R3 were large hos-
pitals with higher demand for RBCs. The demands at these demand points followed
uniform probability distribution on the intervals [5,10], [40,50], and [25,40], respec-
tively. Hence,
⎛ ⎞  ⎛ ⎞ 
 p∈P1 μpx p − 5  p∈P2 μ p x p − 40
P1 ⎝ μp x p⎠ = , P2 ⎝ μp x p⎠ = ,
5 10
p∈P1 p∈P2
⎛ ⎞ 
 p∈P3 μ p x p − 25
P3 ⎝ μp x p⎠ = ,
15
p∈P3

123
Supply chain network operations management of a blood banking system 227

Fig. 3 Supply chain network topology for numerical example 3

where the first O/D pair of nodes is (1, R1 ), the second is (1, R2 ), and the third is
(1, R3 ).
The shortage and outdating penalties for each of the three demand points were:

λ− +
1 = 2200, λ1 = 50,
λ2 = 3000, λ+
−
2 = 60,
λ−
3 = 3000, λ +
3 = 50.

The total risk functions corresponding to the blood collection links were:

r̂1 ( f 1 ) = 2 f 12 , and r̂2 ( f 2 ) = 1.5 f 22 ,

and the risk aversion factor, θ , was 0.7.

The multipliers corresponding to the links, the total cost functions, and the total dis-
carding cost functions were as reported in Table 2. These numbers have been selected
based on the average historical data for the American Red Cross Northeast Division
Blood Services (Rios 2010).
The Euler method (cf. (58)) for the solution of variational inequality (38) was
implemented in Matlab. A Microsoft Windows System with a Dell PC at the Univer-
sity of Massachusetts Amherst was used for all the computations. We set the sequence
aτ = .1(1, 21 , 21 , · · · ), and the convergence tolerance was = 10−6 , that is, the abso-
lute value of the difference between each path flow in two successive iterations was less

123
228 A. Nagurney et al.

Table 2 Total cost and total

Link a αa ĉa ( f a ) ẑ a ( f a ) f a∗
discarding cost functions and
solution for numerical example 3
1 .97 6 f 12 + 15 f 1 .8 f 12 54.72
2 .99 9 f 22 + 11 f 2 .7 f 22 43.90
3 1.00 .7 f 32 + f 3 .6 f 32 30.13
4 .99 1.2 f 42 + f 4 .8 f 42 22.42
5 1.00 f 52 + 3 f 5 .6 f 52 19.57
6 1.00 .8 f 62 + 2 f 6 .8 f 62 23.46
7 .92 2.5 f 72 + 2 f 7 .5 f 72 49.39
8 .96 3 f 82 + 5 f 8 .8 f 82 42.00
9 .98 .8 f 92 + 6 f 9 .4 f 92 43.63
10 1.00 2 +3f
.5 f 10 2
.7 f 10 39.51
10
11 1.00 2 + f
.3 f 11 2
.3 f 11 29.68
11
12 1.00 2 +2f
.5 f 12 2
.4 f 12 13.08
12
13 1.00 2 +2f
.4 f 13 2
.3 f 13 26.20
13
14 1.00 2 + f
.6 f 14 2
.4 f 14 13.31
14
15 1.00 2 +3f
1.3 f 15 2
.7 f 15 5.78
15
16 1.00 2
.8 f 16 + 2 f 16 2
.4 f 16 25.78
17 .98 2 +3f
.5 f 17 2
.5 f 17 24.32
17
18 1.00 2
.7 f 18 + 2 f 18 2
.7 f 18 .29
19 1.00 2 +4f
.6 f 19 2
.4 f 19 18.28
19
20 .98 2
1.1 f 20 + 5 f 20 2
.5 f 20 7.29

than or equal to this . The algorithm was initialized by setting the projected demand
at each demand point and all other variables equal to zero. Table 2 also provides the
computed optimal solutions.
Thus, under the given demand probability distributions for our three demand points,
the amounts of optimal product flow on each link were as reported above. These num-
bers clearly demonstrate the effect of the waste throughout the network, and are subject
to the mentioned link multipliers as well as the various costs associated with those
links, as formerly stated.
The computed amounts of projected demand for each of the three demand points
were:
v1∗ = 6.06, v2∗ = 44.05, and v3∗ = 30.99.

It is interesting to note that, between the two blood collection links, although link 1
has a higher waste/loss rate, and higher total risk and discarding costs, it has a higher
optimal flow of blood product as compared to link 2 due to its lower total cost func-
tion. Furthermore, for the small surgical center, R1 , the value of projected demand,
v1∗ = 6.06, is closer to the lower bound of its uniform probability distribution due to
the relatively smaller shortage penalty cost. In contrast, the values of projected demand
for the larger hospitals, R2 and R3 , are closer to the respective upper bounds of their
uniform distributions.

123
Supply chain network operations management of a blood banking system 229

4 Summary and conclusions

In this paper, we developed a supply chain network optimization model for the man-
agement of the procurement, testing and processing, and distribution of a perishable
product—that of human blood.
The original contributions in the paper, include the blood supply chain network
model operations management model, which has the notable features that it captures
perishability of this life-saving product through the use of arc multipliers; it contains
discarding costs associated with waste/disposal; it handles uncertainty associated with
demand points; it assesses costs associated with shortages/surpluses at the demand
points, and it also quantifies the supply-side risk associated with procurement.
For the sake of generality, and the establishment of the foundations that will enable
further extensions and applications, we used a variational inequality approach for
model formulation and solution. We illustrated the model through transparent numer-
ical examples, which vividly demonstrate the flexibility and generality of our supply
chain network optimization model.
The framework developed here can be applied, with appropriate adaptation, to
other perishable products, such as medicines and vaccines, as well as to agricultural
products, including food.

Acknowledgments This research was supported by the John F. Smith Memorial Fund at the Isenberg
School of Management. This support is gratefully acknowledged. The authors acknowledge Mr. Len Walker,
the Director of Business Development for the American Red Cross Blood Services in the greater Boston
area, who shared valuable information on the subject. They also thank Dr. Jorge Rios, the Medical Director
for the American Red Cross Northeast Division Blood Services, who graciously agreed to be interviewed,
and who provided enlightening thoughts on the model. The first author also acknowledges Professor Georg
Pflug of the University of Vienna, the organizer of the International Conference on Computational Man-
agement Science that took place in Vienna, Austria, July 28–30, 2010 for many interesting conversations
and lectures. Comments and suggestions made by reviewers of earlier versions of this paper were most
appreciated.

References

Ahumada O, Villalobos JR (2009) Application of planning models in the agri-food supply chain: a review.
Eur J Oper Res 195:1–20
Boppana RV, Chalasani S (2007) Analytical models to determine desirable blood acquisition rates. In: IEEE
international conference on system of systems engineering
Cetin E, Sarul LS (2009) blood bank location model: a multiobjective approach. Eur J Pure Appl Math
2(1):112–124
Cohen MA, Pierskalla WP (1979) Target inventory levels for a hospital blood bank or a decentralized
regional blood banking system. Transfusion 19(4):444–454
Dong J, Zhang D, Nagurney A (2004) A supply chain network equilibrium model with random demands.
Eur J Oper Res 156:194–212
Dupuis P, Nagurney A (1993) Dynamical systems and variational inequalities. Ann Oper Res 44:9–42
Federgruen A, Prastacos G, Zipkin P (1986) allocation and distribution model for perishable products. Oper
Res 34(1):75–82
Georgiadis P, Vlachos D, Iakovou E (2005) A system dynamics modeling framework for the strategic
supply chain management of food chains. J Food Eng 70:351–364
Ghandforoush P, Sen TK (2010) DSS to manage platelet production supply chain for regional blood centers.
Decis Support Syst 50(1):32–42

123
230 A. Nagurney et al.

Haijema R (2008) Solving large structured Markov decision problems for perishable—inventory manage-
ment and traffic control. PhD thesis, Tinbergen Institute, The Netherlands
Haijema R, van der Wal J, van Dijk NM (2007) Blood platelet production: optimization by dynamic pro-
gramming and simulation. Comput Oper Res 34:760–779
Hsu CI, Hung SF, Li HC (2007) Vehicle routing problem with time-windows for perishable food delivery.
J Food Eng 80:465–475
Hwang H, Hahn KH (2000) An optimal procurement policy for items with an inventory level-dependent
demand rate and fixed lifetime. Eur J Oper Res 127(3):537–545
Jacobs D, Silan M, Clemson B (1996) An analysis of alternative locations and service areas of American
Red Cross blood facilities. Interfaces 26(3):40–50
Karaesmen IZ, Scheller-Wolf A, Deniz B (2011) Managing perishable and aging inventories: review and
future research directions. In: Kempf KG, Kskinocak P, Uzsoy P (eds) Planning production and
inventories in the extended enterprise. Springer, Berlin pp 393–436
Katsaliaki K, Brailsford SC (2007) Using simulation to improve the blood supply chain. J Oper Res Soc
58:219–227
Mustafee N, Katsaliaki K, Brailsford SC (2009) Facilitating the analysis of a UK national blood service
supply chain using distributed simulation. Simulation 85(2):113–128
Nagurney A (1999) Network economics: a variational inequality approach, second and revised edition.
Kluwer, Dordrecht
Nagurney A (2009) A system-optimization perspective for supply chain network integration: the horizontal
merger case. Transp Res E 45:1–15
Nagurney A (2010a) Supply chain network design under profit maximization and oligopolistic competition.
Transp Res E 46:281–294
Nagurney A (2010b) Optimal supply chain network design and redesign at minimal total cost and with
demand satisfaction. Int J Prod Econ 128:200–208
Nagurney A (2010c) Formulation and analysis of horizontal mergers among oligopolistic firms with insights
into the merger paradox: a supply chain network perspective. Comput Manag Sci 7:377–401
Nagurney A, Aronson J (1989) A general dynamic spatial price network equilibrium model with gains and
losses. Networks 19(7):751–769
Nagurney A, Nagurney LS (2010) Sustainable supply chain network design: a multicriteria perspective. Int
J Sustain Eng 3:189–197
Nagurney A, Qiang Q (2009) Fragile networks: identifying vulnerabilities and synergies in an uncertain
world. Wiley, Hoboken
Nagurney A, Woolley T (2010) Environmental and cost synergy in supply chain network integration in merg-
ers and acquisitions. In: Ehrgott M, Naujoks B, Stewart T, Wallenius J (eds) Sustainable energy and
transportation systems. Proceedings of the 19th international conference on multiple criteria decision
making. Lecture notes in economics and mathematical systems. Springer, Berlin, pp 51–78
Nagurney A, Zhang D (1996) Projected dynamical systems and variational inequalities with applications.
Kluwer, Boston
Nagurney A, Cruz J, Dong J, Zhang D (2005) Supply chain networks, electronic commerce, and supply
side and demand side risk. Eur J Oper Res 164:120–142
Nagurney A, Yu M, Qiang Q (2011) Supply chain network design for critical needs with outsourcing. Papers
Reg Sci 90(1):123–143
Nahmias S (1982) Perishable inventory theory: a review. Oper Res 30(4):680–708
Omosigho SE (2002) Determination of outdate and shortage quantities in the inventory problem with fixed
lifetime. Int J Comput Math 79(11):1169–1177
Osvald A, Stirn L (2008) A vehicle routing algorithm for the distribution of fresh vegetables and similar
perishable food. J Food Eng 85:285–295
Pierskalla WP (2004) Supply chain management of blood banks. In: Brandeau ML, Sanfort F, Pierskalla
WP (eds) Operations research and health care: a handbook of methods and applications. Kluwer,
Boston pp 103–145
Prastacos GP (1984) Blood inventory management: an overview of theory and practice. Manag Sci
30(7):777–800
Redcrossblood.org. (2010) Donation FAQs. http://www.redcrossblood.org/donating-blood/donation-faqs
Rios J (2010) Interview with the Medical Director for the American Red Cross Northeast Division Blood
Services. Dedham, Massachusetts on July 19, 2010

123
Supply chain network operations management of a blood banking system 231

Rytila JS, Spens KM (2006) Using simulation to increase efficiency in blood supply chains. Manag Res
News 29(12):801–819
Sahin G, Sural H, Meral S (2007) Locational analysis for regionalization of Turkish Red Crescent blood
services. Comput Oper Res 34:692–704
Sivakumar P, Ganesh K, Parthiban P (2008) Multi-phase composite analytical model for integrated alloca-
tion-routing problem—application of blood bank logistics. Int J Logist Econ Global 1(3/4):251–281
Sullivan M, Cotten R, Read E, Wallace E (2007) Blood collection and transfusion in the United States in
2001. Transfusion 47:385–394
The New York Times (2010) In a world of throwaways, making a dent in medical waste. Chen, I., July 5.
http://www.nytimes.com/2010/07/06/health/06waste.html
Walker L (2010) Phone interview with the Director of Business Development for the American Red Cross
Blood Services in the Greater Boston Area on July 14
Whitaker BI, Green J, King MR, Leibeg LL, Mathew SM, Schlumpf KS, Schreiber GB (2007) The 2007
national blood collection and utilization survey report. The United States Department of Health and
Human Services
Yang X (2006) Choosing transportation alternatives for highly perishable goods: a case study on nuclear
medicine. Masters thesis, Massachusetts Institute of Technology, Cambridge
Yegul M (2007) Simulation analysis of the blood supply chain and a case study. Master’s thesis, Middle
East Technical University, Turkey
Zanoni S, Zavanella L (2007) Single-vendor single-buyer with integrated transport-inventory system: Mod-
els and heuristics in the case of perishable goods. Comput Ind Eng 52:107–123
Zhou YW, Yang SL (2003) An optimal replenishment policy for items with inventory-level-dependent
demand and fixed lifetime under the LIFO policy. J Oper Res Soc 54(6):585–593

123